Skip to content
← Back to explorer

HFEPX Hub

Automatic Metrics + Math (Last 90 Days)

Updated from current HFEPX corpus (Mar 8, 2026). 13 papers are grouped in this hub page.

Read Full Context

Updated from current HFEPX corpus (Mar 8, 2026). 13 papers are grouped in this hub page. Common evaluation modes: Automatic Metrics, Llm As Judge. Most common rater population: Domain Experts. Common annotation unit: Trajectory. Frequently cited benchmark: Bankmathbench. Common metric signal: accuracy. Use this page to compare protocol setup, judge behavior, and labeling design decisions before running new eval experiments. Newest paper in this set is from Feb 25, 2026.

Papers: 13 Last published: Feb 25, 2026 Global RSS Tag RSS
Automatic MetricsMathLast 90d

Researcher Quick Triage

This hub is best used for protocol triage and replication planning from abstract-level evidence. Quality band: Developing .

All Sampled Papers (13) Replication-Ready Only (2)

High-Signal Coverage

100.0%

13 / 13 sampled papers are not low-signal flagged.

Replication-Ready Set

2

Benchmark + metric + eval mode explicitly present.

Judge/Human Comparability

0

Papers containing both `human_eval` and `llm_as_judge`.

  • 2 papers are replication-ready (benchmark + metric + explicit evaluation mode).
  • 0 papers support judge-vs-human agreement analysis.
  • 0 papers report explicit quality controls (calibration/adjudication/IAA).

Primary action: Use this page for scouting only; collect additional papers before attempting replication-critical comparisons.

Currently showing only replication-ready papers in ranking and matrix sections (2 papers).

Why This Matters For Eval Research

  • 15.4% of papers report explicit human-feedback signals, led by pairwise preferences.
  • automatic metrics appears in 100% of papers in this hub.
  • Bankmathbench is a recurring benchmark anchor for cross-paper comparisons in this page.

Protocol Takeaways

  • Quality-control reporting is sparse in this slice; prioritize papers with explicit calibration or adjudication steps.
  • Rater context is mostly domain experts, and annotation is commonly trajectory-level annotation; use this to scope replication staffing.
  • Pair this hub with a human_eval-heavy hub to validate judge-model calibration.

Benchmark Interpretation

  • Bankmathbench appears in 7.7% of hub papers (1/13); use this cohort for benchmark-matched comparisons.
  • LiveCodeBench appears in 7.7% of hub papers (1/13); use this cohort for benchmark-matched comparisons.

Metric Interpretation

  • accuracy is reported in 69.2% of hub papers (9/13); compare with a secondary metric before ranking methods.
  • cost is reported in 23.1% of hub papers (3/13); compare with a secondary metric before ranking methods.
Researcher Checklist (Expanded)

Researcher Checklist

  • Gap: Papers with explicit human feedback

    Coverage is a replication risk (15.4% vs 45% target).

  • Gap: Papers reporting quality controls

    Coverage is a replication risk (0% vs 30% target).

  • Moderate: Papers naming benchmarks/datasets

    Coverage is usable but incomplete (23.1% vs 35% target).

  • Strong: Papers naming evaluation metrics

    Coverage is strong (92.3% vs 35% target).

  • Gap: Papers with known rater population

    Coverage is a replication risk (7.7% vs 35% target).

  • Strong: Papers with known annotation unit

    Coverage is strong (38.5% vs 35% target).

Strengths

  • Agentic evaluation appears in 76.9% of papers.

Known Gaps

  • Only 0% of papers report quality controls; prioritize calibration/adjudication evidence.
  • Rater population is under-specified (7.7% coverage).
  • LLM-as-judge appears without enough inter-annotator agreement reporting.

Suggested Next Analyses

  • Pair this hub with a human_eval-heavy hub to validate judge-model calibration.
  • Stratify by benchmark (Bankmathbench vs LiveCodeBench) before comparing methods.
  • Track metric sensitivity by reporting both accuracy and cost.
Recommended Queries (Expanded)

Recommended Queries

LLM-as-Judge Protocols Benchmark Slice: Bankmathbench Metric Slice: accuracy Recent High-Signal Papers

Start with These 3

Use these when you need one protocol anchor, one benchmark anchor, and one recent comparison point before reading the wider hub.

Start Here (Best First 6)

Ranked for protocol completeness (human signal, benchmark + metric anchors, quality controls, and judge/human overlap).

Protocol Matrix (Top 12)

Use this to quickly compare protocol ingredients instead of scanning long prose.

Paper HF Signal Eval Modes Benchmarks Metrics QC
Duel-Evolve: Reward-Free Test-Time Scaling via LLM Self-Preferences

Feb 25, 2026

 Yes
Pairwise Preference
Automatic Metrics LiveCodeBench , Mathbench Accuracy Not Reported
BankMathBench: A Benchmark for Numerical Reasoning in Banking Scenarios

Feb 19, 2026

 No
Not Reported
Automatic Metrics Bankmathbench Accuracy Not Reported

Protocol Diff (Top Papers)

Fast side-by-side comparison for the highest-ranked papers in this hub.

Signal Duel-Evolve: Reward-Free Test-Time Scaling via LLM… BankMathBench: A Benchmark for Numerical Reasoning…
Human Feedback Pairwise PreferenceNot reported
Evaluation Modes Automatic MetricsAutomatic Metrics
Benchmarks LiveCodeBench, MathbenchBankmathbench
Metrics AccuracyAccuracy
Quality Controls Not reportedNot reported
Rater Population UnknownUnknown
Annotation Unit PairwiseUnknown
Suggested Reading Order (Extended)

This section is intentionally expanded only when needed; use “Start Here” above for a faster pass.

Suggested Reading Order

  1. Replaying pre-training data improves fine-tuning

    Start here for detailed protocol reporting and quality-control evidence. Signals: automatic metrics. Focus: accuracy. Abstract: To obtain a language model for a target domain (e.g.

  2. GenDB: The Next Generation of Query Processing -- Synthesized, Not Engineered

    Start here for detailed protocol reporting and quality-control evidence. Signals: automatic metrics. Focus: cost. Abstract: Traditional query processing relies on engines that are carefully optimized and engineered by.

  3. Surgical Post-Training: Cutting Errors, Keeping Knowledge

    Start here for detailed protocol reporting and quality-control evidence. Signals: automatic metrics + pairwise preferences. Focus: accuracy. Abstract: While prior research emphasizes the role of on-policy data in.

  4. Duel-Evolve: Reward-Free Test-Time Scaling via LLM Self-Preferences

    Include a human-eval paper to calibrate against judge-based evaluation settings. Signals: automatic metrics + pairwise preferences. Focus: LiveCodeBench / accuracy. Abstract: Pairwise comparisons, by contrast, are often easier.

  5. Gradient Regularization Prevents Reward Hacking in Reinforcement Learning from Human Feedback and Verifiable Rewards

    Include an LLM-as-judge paper to test judge design and agreement assumptions. Signals: LLM-as-judge. Focus: accuracy. Abstract: GR achieves a higher GPT-judged win-rate in RLHF, avoids overly focusing on.

  6. BankMathBench: A Benchmark for Numerical Reasoning in Banking Scenarios

    Adds automatic metrics for broader protocol coverage within this hub. Signals: automatic metrics. Focus: Bankmathbench / accuracy. Abstract: Large language models (LLMs)-based chatbots are increasingly being adopted in.

  7. Recycling Failures: Salvaging Exploration in RLVR via Fine-Grained Off-Policy Guidance

    Adds automatic metrics for broader protocol coverage within this hub. Signals: automatic metrics. Focus: accuracy. Abstract: Reinforcement Learning from Verifiable Rewards (RLVR) has emerged as a powerful paradigm.

  8. Test-Time Scaling with Diffusion Language Models via Reward-Guided Stitching

    Adds automatic metrics for broader protocol coverage within this hub. Signals: automatic metrics. Focus: accuracy. Abstract: Reasoning with large language models often benefits from generating multiple chains-of-thought, but.

Known Limitations

Known Limitations

  • Only 0% of papers report quality controls; prioritize calibration/adjudication evidence.
  • Rater population is under-specified (7.7% coverage).
  • Narrative synthesis is grounded in metadata and abstracts only; full-paper implementation details are not parsed.
Research Utility Snapshot (Detailed)

Research Utility Snapshot

Human Feedback Mix

  • Pairwise Preference (2)

Evaluation Modes

  • Automatic Metrics (13)
  • Llm As Judge (1)

Top Benchmarks

  • Bankmathbench (1)
  • LiveCodeBench (1)
  • MATH (1)
  • Mathbench (1)

Top Metrics

  • Accuracy (9)
  • Cost (3)
  • Agreement (1)
  • Coherence (1)

Rater Population Mix

  • Domain Experts (1)

Quality Controls

Coverage diagnostics (sample-based): human-feedback 23.1% · benchmarks 23.1% · metrics 92.3% · quality controls 0.0%.

Top Papers

Related Hubs

Need human evaluators for your AI research? Scale annotation with expert AI Trainers.

Post a Job Get a Quote